Operations Excellence

Designing and provisioning a system is day-1 work; keeping it healthy in production is day 2, and that is what Operations Excellence covers. The mental model that ties the whole domain together comes from Google's site reliability engineering (SRE) practice: pick a service level objective (SLO), a reliability target deliberately set below 100% (say 99.9%), and treat the leftover (1 minus the SLO) as an error budget you are allowed to spend on change. Reliability stops being a vibe and becomes a number you measure, alert on, and trade against release velocity. The classic exam trap this dodges is the instinct to chase 100% uptime: the right answer reflects an explicit SLO with an error budget, because perfect reliability is neither achievable nor worth its cost. This domain maps to two pillars of the Google Cloud Well-Architected Framework, operational excellence (how you operate) and reliability (how you prove the system survives failure), so a question can come at you from either angle.

Read the subtopics as one running loop. Operational Excellence Pillar is the framing: the five core principles of the Well-Architected operational excellence pillar and the SRE vocabulary (SLO, SLA, SLI, error budget, toil, blameless postmortem) that the rest of the domain reuses. Cloud Observability is how you see the system: the suite formerly named Stackdriver, with Cloud Monitoring, Cloud Logging, Cloud Trace, Cloud Profiler, and Error Reporting, plus where you alert on the four golden signals. Release Management is how you ship safely: a Cloud Deploy delivery pipeline, canary rollouts, approvals, and one-command rollback. Production Support is the support you buy (Customer Care tiers, case priorities) and the incident practice you run (incident versus problem management, on-call). Quality Control is the set of gates that decide whether work may advance, and the DORA metrics that score the delivery system. Production Reliability is how you prove resilience by testing it, through chaos engineering, load and stress testing, DR drills, and penetration tests.

Operations Excellence has a consistent instinct across all six subtopics: prefer the choice that removes human toil and limits blast radius. A gate that blocks beats a report that merely warns; an automated canary with a metric check beats a human watching a dashboard; a one-command rollback beats a hot-patch in production; a blameless postmortem that fixes the systemic cause beats faster firefighting; infrastructure as code under version control beats a console click. Toil, the repetitive manual work that grows with the service, is the thing you are always trying to automate away. When two options are both technically correct, pick the one that codifies the operation so it survives the next 3 a.m. page.